Credit Card Fraud Detection using XGBoost

Project Overview

This project focuses on detecting fraudulent credit card transactions using a machine learning pipeline. Transaction data includes merchant details, amount, user demographics, and time-based features. The model is built using XGBoost, which is highly effective for handling imbalanced tabular datasets.

---

Task Objectives

• Build a fraud detection system that minimizes false positives while maximizing fraud detection accuracy.
• Perform EDA (Exploratory Data Analysis) to identify patterns in fraudulent transactions.
• Preprocess the data using standard techniques and OneHotEncoding for categorical features.
• Train an XGBoost Classifier with engineered features.
• Evaluate the model using metrics like confusion matrix, ROC-AUC, and classification report.
• Analyze misclassifications and their implications.

---

Project Structure

Credit-Card-Fraud-Detection/
│
├── credit_card_fraud_detection_xgboost.ipynb   # Main Colab notebook
├── EDA_Results/                                # Folder containing EDA result plots
│   ├── fraud_rate.png
│   ├── transaction_amount_distribution.png
├── output/                                     # Model evaluation visualizations
│   ├── precision_recall_curve.png
│   ├── roc_curve.png
│   ├── confusion_matrix.png
│
└── README.md                                   # Project documentation

⚠️ Note: Dataset files (fraudTrain.csv and fraudTest.csv) are not included due to size restrictions. Please download them from Kaggle.

---

Steps to Run the Project in Google Colab

1. Open the Colab Notebook

• Download or copy the notebook credit_card_fraud_detection_xgboost_colab.ipynb to your Google Drive.
• Open it with Google Colab.

2. Install Dependencies

At the top of the Colab notebook, run the following cell to install required packages:

pip install pandas matplotlib seaborn scikit-learn xgboost

3. Upload the Dataset

• Download the dataset files and place them in the root directory:
  • fraudTrain.csv
  • fraudTest.csv
• In Colab, use the file upload widget to upload both files:

  from google.colab import files
  uploaded = files.upload()

4. Run the Notebook

Execute all cells in order to perform EDA, preprocessing, model training, and evaluation.

---

📊 Output Highlights

• EDA Visualizations:
  • Fraud rate by transaction category
  • Transaction amount distribution for fraud vs. non-fraud
• Model Evaluation:
  • Accuracy, precision, recall, F1-score
  • Confusion matrix (with heatmap)
  • ROC curve and AUC score

---

⚠️ Misclassification Insights

• False Positives (FP): Legitimate transactions flagged as fraud – user inconvenience.
• False Negatives (FN): Fraudulent transactions missed – potential financial loss.
• Focus is placed on minimizing FN while keeping FP under control.

---

📌 Notes

• Features include age, transaction hour, and day of week, derived from datetime fields.
• Data is preprocessed using StandardScaler for numeric and OneHotEncoder for categorical features.
• The model is trained on fraudTrain.csv and tested on fraudTest.csv.